Christopher D. Rahn

Soft robotics: Biological inspiration, state of the art, and future research

Traditional robots have rigid underlying structures that limit their ability to interact with their environment. For example, conventional robot manipulators have rigid links and can manipulate objects using only their specialised end effectors. These robots often encounter difficulties operating in unstructured and highly congested environments. A variety of animals and plants exhibit complex movement with soft structures devoid of rigid components. Muscular hydrostats e.g. octopus arms and elephant trunks are almost entirely composed of muscle and connective tissue and plant cells can change shape when pressurised by osmosis. Researchers have been inspired by biology to design and build soft robots. With a soft structure and redundant degrees of freedom, these robots can be used for delicate tasks in cluttered and/or unstructured environments. This paper discusses the novel capabilities of soft robots, describes examples from nature that provide biological inspiration, surveys the state of the art and outlines existing challenges in soft robot design, modelling, fabrication and control.

Model-Based Electrochemical Estimation and Constraint Management for Pulse Operation of Lithium Ion Batteries

High-power lithium ion batteries are often rated with multiple current and voltage limits depending on the duration of the pulse event. These variable control limits, however, are difficult to realize in practice. In this paper, a linear Kalman filter based on a reduced order electrochemical model is designed to estimate internal battery potentials, concentration gradients, and state-of-charge (SOC) from external current and voltage measurements. A reference current governor predicts the operating margin with respect to electrode side reactions and surface depletion/saturation conditions responsible for damage and sudden loss of power. The estimates are compared with results from an experimentally validated, 1-D, nonlinear finite volume model of a 6 Ah hybrid electric vehicle battery. The linear filter provides, to within ~ 2%, performance in the 30%-70% SOC range except in the case of severe current pulses that draw electrode surface concentrations to near saturation and depletion, although the estimates recover as concentration gradients relax. With 4 to 7 states, the filter has low-order comparable to empirical equivalent circuit models commonly employed and described in the literature. Accurate estimation of the batterys internal electrochemical state enables an expanded range of pulse operation.

Geometrically Exact Models for Soft Robotic Manipulators

Unlike traditional rigid linked robots, soft robotic manipulators can bend into a wide variety of complex shapes due to control inputs and gravitational loading. This paper presents a new approach for modeling soft robotic manipulators that incorporates the effect of material nonlinearities and distributed weight and payload. The model is geometrically exact for the large curvature, shear, torsion, and extension that often occur in these manipulators. The model is based on the geometrically exact Cosserat rod theory and a fiber reinforced model of the air muscle actuators. The model is validated experimentally on the OctArm V manipulator, showing less than 5% average error for a wide range of actuation pressures and base orientations as compared to almost 50% average error for the constant-curvature model previously used by researchers. Workspace plots generated from the model show the significant effects of self-weight on OctArm V.

Design of Continuous Backbone, Cable-Driven Robots

Continuous backbone robots driven by cables have many potential applications in dexterous manipulation for manufacturing and space environments. Design of these robots requires specification of a stiff yet bendable backbone, selection of cable support heights and spacings, and development of a cable drive system. The robot arm divides into sections that are subdivided into segments bounded by cable supports. Cable pairs attach to the end of each section and provide two axis bending. Thus, with many sections, the arm can be bent into complex shapes to allow redundant positioning of the end effector payload. The kinematics of the entire arm are determined from the segment kinematics. This paper derives and numerically solves the nonlinear kinematics for a single segment of a continuous backbone robot. Optimal spacing of the cable supports maximizes displacement, load capacity, and simplicity of the robot kinematics. An experimental system verifies the theoretically predicted performance.

Variable Stiffness Structures Utilizing Fluidic Flexible Matrix Composites

In this research, the capability of utilizing fluidic flexible matrix composites (F2MC) for autonomous structural tailoring is investigated. By taking advantage of the high anisotropy of flexible matrix composite (FMC) tubes and the high bulk modulus of the pressurizing fluid, significant changes in the effective modulus of elasticity can be achieved by controlling the inlet valve to the fluid-filled F2MC structure. The variable modulus F2MC structure has the flexibility to easily deform when desired (open-valve), possesses the high modulus required during loading conditions when deformation is not desired (closed-valve — locked state), and has the adaptability to vary the modulus between the flexible/stiff states through control of the valve. In the current study, a 3D analytical model is developed to characterize the axial stiffness behavior of a single F 2MC tube. Experiments are conducted to validate the proposed model, and the test results show good agreement with the model predictions. A closed/open modulus ratio as high as 56 times is achieved experimentally. With the validated model, an F2MC design space study is performed. It is found that by tailoring the properties of the FMC tube and inner liner, a wide range of moduli and modulus ratios can be attained. By embedding multiple F 2MC tubes side by side in a soft matrix, a multi-cellular F2MC sheet with a variable stiffness in one direction is constructed. The stiffness ratio of the multi-cellular F2MC sheet obtained experimentally shows good agreement with a model developed for this type of structure. A case study has been conducted to investigate the behavior of laminated [+60/0/-60] s multi-cellular F2MC sheets. It is shown that the laminate can achieve tunable, steerable, anisotropy by selective valve control.

Adaptive vibration isolation for axially moving strings: theory and experiment

High-speed transport of continuous materials such as belts, webs, filaments, or bands can cause unwanted vibration. Vibration control for these systems often focuses on restricting the response resulting from external disturbances (e.g. support roller eccentricity or aerodynamic excitation) to areas not requiring high precision positioning. This paper introduces vibration controllers for an axially moving string system consisting of a controlled span coupled to a disturbed span via an actuator. The system model includes a partial differential equation for the two spans and an ordinary differential equation for the actuator. Exact model knowledge and adaptive isolation controllers, based on Lyapunov theory, regulate the controlled span from bounded disturbances in the adjacent, uncontrolled span. Assuming distributed damping in the uncontrolled span, the exact model knowledge and adaptive controllers exponentially and asymptotically drive the controlled span displacement to zero, respectively, while ensuring bounded uncontrolled span displacement and control force. Experiments demonstrate the effectiveness of the proposed controller in isolating the controlled span from disturbances and damping the controlled span displacement.

Model Order Reduction of 1D Diffusion Systems Via Residue Grouping

A model order reduction method is developed and applied to 1D diffusion systems with negative real eigenvalues. Spatially distributed residues are found either analytically (from a transcendental transfer function) or numerically (from a finite element or finite difference state space model), and residues with similar eigenvalues are grouped together to reduce the model order. Two examples are presented from a model of a lithium ion electrochemical cell. Reduced order grouped models are compared to full order models and models of the same order in which optimal eigenvalues and residues are found numerically. The grouped models give near-optimal performance with roughly 1/20 the computation time of the full order models and require 1000―5000 times less CPU time for numerical identification compared to the optimization procedure.

Profile sensing with an actuated whisker

Obstacle avoidance and object identification are important tasks for robots in unstructured environments. This paper develops an actuated whisker that determines contacted object profiles using a hub load cell. The shape calculation algorithm numerically integrates the elastica equations from the measured hub angle, displacement, forces, and torque until the bending moment vanishes, indicating the contact point. Sweeping the whisker across the object generates a locus of contact points that can be used for object identification. Experimental results demonstrate the ability to identify and differentiate square and curved objects at various orientations.

Design of an artificial muscle continuum robot

This paper introduces a novel continuum manipulator consisting of two flexible sections connected by rigid base plates. Each section uses 6-8 opposing contracting and extending McKibben actuators to provide two-axis bending. The contracting/extending actuator pairs are sized and positioned to provide matched rotation and torque. The experimental manipulator generates large rotation (30 - 35 degrees per section and 50 degrees for the entire manipulator) and load capacity (25 lb vertical lift) at 60 psi. With a large strength to weight ratio, this manipulator also is highly flexible, simple to manufacture, and is cost-effective.

Three-dimensional contact imaging with an actuated whisker

Contact sensors can provide high-information-density object surface sensing in harsh and/or opaque environments. This paper describes the design, modeling, control, and data processing of a contact imager consisting of a flexible whisker mounted on a two-axis robot through a load cell. The whisker sweeps around and into contact with unknown objects, determining the three-dimensional location of contact points to within a specified position resolution. During contact, the whisker bends along the surface normal, producing large deflections. The joint angles and load cell signals are numerically processed to determine the whisker shapes. Comparison of whisker shapes during bending determines contact point location. Experimental results for several objects with wide ranging surface curvature and roughness demonstrate 1.51-cm resolution for a 45.5-cm whisker